One of the important issues in intelligent systems and robotics is to develop an efficient method to control multi-agent system. In order to work the multi-agent system well as the problem solver, it's so significant to create cooperative behaviors among the agents. In the multi-agent system, the behaviors of cooperation emerged as the results of suitable position learning by each agent. In this paper, we design dynamic positioning method based on the Neural network and Competitive co-evolution model further more We applied this method to Soccer Server.